Health Informatics in Neonatology: What is the Potential to Further Improve Outcomes?

• Abstract
• Background
• Neonatology
• Impact of Data on Clinical Care
• Knowledge Discovery
• Outcome Prediction
• Role of Data From Neonatal Networks
• Use of New Technologies
• Conclusion
• References
• Acknowledgements

Abstract
Neonatology is the subspecialty of paediatrics that provides care to the smallest and sickest of newborn infants. This type of care generates large amounts of data, which might include demographic details, clinical coding and the results of monitoring or investigations. This paper explores the potential merit of and possible barriers to the use of an informatics approach to the analysis of this data with the aim of maximising the potential health gains.

Background
Neonatology is a medical speciality that provides significant opportunities for advanced data collection and analysis methods. Clearly current informatics techniques have much to offer in this regard. In this paper I will review, in turn, the potential merit of and possible barriers to novel approaches to the analysis of health data in the field of neonatology. I will start by briefly outlining what is involved in this field of health care and then review the potential data sources and uses.

Neonatology
Neonatology is a subspecialty of paediatrics that consists of the medical care of newborn infants, especially the ill or premature newborn infant.^[1] Neonatal intensive care, or Level III care, is a service for the smallest and sickest of newborn infants and is provided at a small number of specialised centres. Level II care provides for less severe neonatal problems and is provided by a greater number of centres. Local guidelines exist for assessment and management of the newborn on admission to neonatal intensive care units (NICUs).^[2] This type of care can generate large amounts of data, such as: demographic data, including maternal age and ethnicity; antenatal clinical data, including pregnancy complications; neonatal, ie, infant data, such as gestation, birth weight, Apgar scores and treatment details. Such infants are monitored closely and recordings of blood pressure, heart rate, oxygen saturation and blood gas analysis are collected at regular intervals forming a sizeable volume of data.

In addition to the amount of data generated, there are some characteristics of the speciality that make the analysis particularly pertinent. Neonatology is similar to other intensive care specialities in that it provides care to seriously unwell individuals, the clinical condition of the patients may rapidly change and clinical decisions are frequently based on a mixture of clinical and monitoring data. However, there are a number of differences between neonatal and adult or paediatric intensive care.

Typically, neonatal care involves a more homogeneous population with a narrower range of presenting conditions, greater patient numbers and a longer mean stay in the unit. Indeed, prematurity, per se, would be the most common indication for neonatal intensive care. These characteristics tend to increase the utility of a "population" approach to data analysis.

In addition, neonatologists utilise data analyses extensively in their day-to-day practice and to assist in patient management have created scoring systems such as CRIB, CRIB II and SNAP.^[3-5] These may be used to predict outcome or in benchmarking exercises.

Impact of Data on Clinical Care
The potential for using well-formatted and accessible data to provide useful information to guide clinical care for individual infants is obvious. Blood gas data would influence ventilation changes, antibiotic changes would be directed by microbiological sensitivity data and anticonvulsant use would be indicated by the presence of clinical seizures or EEG changes. While there have been interesting attempts to provide real time "expert systems" to guide neonatal care,^[6-8] the formal, structured and ongoing collection of the background data has not been well described, particularly when the use of or method of analysis of the data can only be decided in the future as patterns of disease and correlations between cause and effect begin to emerge. This difficult area is the focus of the current paper.

At the most prosaic level, data could be collected to a simple spreadsheet or database then analyses could be performed using Microsoft Excel or a similar program. Examples of these simple systems could include sepsis surveillance where data may be collected on infection rates, organisms and antibiotic sensitivity. However more advanced systems have the potential to offer more inferential power and, for example, could also collect data on associated morbidities and/or risk factors such as duration of in-dwelling catheters. As the system becomes more sophisticated the data collection could be prompted by a computer at the point of care or collected automatically from laboratory systems. From a population perspective, reports could be automated and include analysis of changes in organisms such as rates of resistant organism or emerging organisms. An example of this type of surveillance might be examination of changes in the rates of early onset sepsis with ampicillin-resistant Escherichia coli following the wide spread introduction of antibiotic prophylaxis for newborns at risk of group B streptococcus infection.^[9,10]

Systematic data collection and analysis is also vital in delineating the perinatal effects of emerging infections.^[11,12] In its most advanced form a system might offer complex functions such as syndrome surveillance.^[13] This would involve timely, in theory even real time, analysis that would automatically provide interpretable information that included outbreak verification as well as following the size and spread. Although such systems should not replace traditional public health surveillance or direct reporting of conditions of public health importance, they could provide a timely head start that might have a role in preparation and public health management of the condition.^[14]

Knowledge Discovery
Other exciting developments come from novel methods of knowledge discovery and analysis. Under this heading may be included data mining, neural networks and artificial intelligence systems.

Data mining is considered to have many aspects,^[15] including the discovery of unsuspected relationships that are of value to the database owners and prediction of future events or, more commonly in clinical care, patient outcomes.

The use of neural networks offers particular promise in this regard due to the potential to detect patterns or trends that are non-linear and frequently too complex to be identified by other means. Neural networks are computer systems that operate in a manner that is analogous to the brain; a "neuron" receives and processes input signals then, depending on the result of processing, decides whether or not to stimulate the many other "neurons" to which it is connected. It is able to perform sophisticated modelling that is appropriate for complex medical conditions. The last 15 years have seen a massive increase in the use of neural networks and they are now used in a vast number of automated processes, from voice recognition through economic forecasting to biotechnology applications. The potential in the clinical area, particularly intensive care, was recognised over 10 years ago.^[16] In paediatric intensive care, neural networks may predict length of stay and acuity of care as early as 10 minutes after presentation.^[17] Such information is valuable for both resource planning and patient management. Despite the theoretical advantages of neural networks for analysis there are some technical limitations^[18] and difficulties in their clinical use.^[19] A review of papers comparing neural networks with regression reported similar performance for the techniques in 50 percent of the papers, neural networks performing better than regression in 36 percent and visa versa in 14 percent.^[19] However, it should be noted that in some studies the neural network could potentially have been limited by only having access to data that "experts" considered useful; and in some studies neither technique yielded sufficient power to be used clinically. Also problematic is that each measurement has an error, which may contribute to problems in complex analysis. Finally, it might not be easy to interpret the output from a neural network in a clinical context. In a regression model inferences can easily be drawn, so it has been suggested that neural networks and regression should be used in a complementary manner.^[19] Although this seems reasonable a recent review suggests that "it may be possible to shed light on the black box",^[20] which would allow greater interpretation of the outputs. From a clinical perspective, the ability to improve output interpretation would lead to a further increase in utility of neural networks.

The process of knowledge discovery will be considered further at a later point with respect to both research and quality applications; meanwhile, the second utility of forecasting has considerable attraction in neonatology. To date the published literature has focused on two central questions. The prediction of preterm birth^[21] and the prediction of outcome, including mortality, morbidity and service utilisation, after preterm birth has occurred.^[22-27] In the majority of studies the system has "learnt" using data representative of a part of the cohort, then has been used to predict outcomes in the remaining cohorts. With regard to predicting preterm birth, a recent study reported an artificial neural network had a sensitivity of 54.8 percent;^[21] although encouraging, this still requires further development. In New Zealand, the SCOPE study^[28] plans to examine a large number of women during pregnancy and utilise a number of clinical and molecular markers to predict pregnancy complications including preterm delivery and growth restriction. Advanced data management will be required to prospectively explore combinations of these molecular and clinical markers.

Outcome Prediction
Once an infant is born preterm, standard regression analysis or an artificial intelligence approach may assist with outcome prediction.^{[23, 25, 29]} Although it should be noted that standard clinical predictors such as birth weight, gestation and Apgar score still account for much of the predictive value, as measured by area under the curve.^[24] Service utilisation, measured by length of stay,^[26] and morbidity such as intraventricular haemorrhage may be predicted using either neural networks or logistic regression with the area under the Receiver Operator Curve being superior for the neural network method.^[27] Nevertheless, it has been noted that an artificial neural network, even when trained on admission data and predicting mortality risk for the majority of preterm infants, still has a small but significant number of prediction failures and, thus, may be unsuitable for individual treatment decisions.^[29] In practice, decisions about the withdrawal of intensive care are made by a clinician, who is aware of individual clinical factors, in conjunction with the family; but robust estimates of the risk of mortality or adverse outcome would be fundamental to this process.

On the other hand, some clinical decisions are not final or life threatening and prediction based on a percentage or proportion would be useful. An example of this is a prediction of success in extubation (coming off a ventilator). About a third of intubated preterm infants will fail at extubation and so require repeat intubation, often with sedation and paralysis. After repeat ventilation such an infant will likely require more support due to atelectasis of the lungs. Assistance with prediction could decrease significantly the reintubation rates, so avoiding both a procedure for the infant and cost for the service. Current work suggests a potential for using data in this way to provide a decision-support tool.^[30]

Role of Data From Neonatal Networks
The second major use of analysis of aggregate data is for comparison between centres or services for the purposes of examining quality of the care provided. Many neonatal units now belong to neonatal networks – defined as a collaboration involving more than one clinical site where a common protocol is used for a randomised trial, observational study, or quality improvement project.^[31] The centre regularly submits data, which is then collated and presented in a format that allows comparison between the whole network and the individual units. This type of data could be used to examine variation in both clinical practice and outcomes.^[32] Furthermore, it could be used to facilitate service improvement.^[33] A common outcome to study is intraventricular haemorrhage.^[32,34] This is a potentially serious complication of preterm birth that can have lifelong adverse neurological consequences. The incidence might relate to some aspects of clinical care but is also dependent on case mix so adjustment must be made for this. Published studies suggest there is some variation across units that may be dependent on NICU characteristics,^[34] specifically the NICUs with high patient volume and high neonatologist/staff ratio had lower rates of severe intraventricular haemorrhage. However, attempting to shift all centres towards the better centres may have more effect overall than concentrating on the worst performing hospitals.^[32] Moreover, there is some concern about the validity of creating league tables using data that has not been not fully adjusted or recorded over only a limited time period.^[35]

Another advantage of a network is that the combined data from the whole network provides a greater capability to examine important questions on the quality of clinical care. Such topics would include reasons for variation in rates of sepsis where network data demonstrated that some types of intravenous lines were a major risk factor.^[36] Other important network based studies include comparison of outcome by hospital type^[37] and mortality rates for infants admitted to neonatal intensive care units at night.^[38] These studies demonstrated that outborn infants (ie, those born at one centre, then transferred after birth to receive specialist care) admitted to paediatric hospitals were at higher risk of death, nosocomial infection and oxygen dependency at 28 days of age than those admitted to perinatal centres and that neonatal mortality was higher among inborn infants admitted at night. These findings have major implications for service planning.

A further benefit of these large collations of data generated in a neonatal network is that it can be interrogated to yield important information on the role of specific risk factors. These risk factors may in turn be modified (if possible) to improve outcomes. Examples of such outcomes studied in this manner include intraventricular haemorrhage,^[39] chronic lung disease,^[40] retinopathy of prematurity^[41] and death.^[42] Unfortunately, the major risk factors identified include gestation, growth restriction and male sex, all of which have hitherto proven hard to modify. However, if an automated analysis, not constrained by conventional medical thinking, had the opportunity to explore the data it might have been able to demonstrate somewhat earlier that treatment with erythropoietin carried an increased risk of retinopathy of prematurity. This information has only recently come to light following meta analysis of several trials.^[43] Indeed, a similar approach could be applied to congenital anomalies, with each case reported to a central registry.^[44] Ideally, such a system could link with other data sources that would include maternal clinical data and medications and potentially provide early warnings and decrease the risk of a tragedy similar to that of thalidomide.^[45] Further, data-mining techniques can be used in disease specific research to examine a broad range of variables that might include both clinical and biochemical data. Examples include the study of surfactant genes^[46] in bronchopulmonary dysplasia and perioperative study of congenital cardiac disease.^[47]

Use of New Technologies
Another way that shared network data may be utilised is as an assessment of trends in the use of new technologies. Examples of this include a study of the use of high frequency ventilation in Australia and New Zealand.^[48] Data were available from 3270 infants receiving high frequency ventilation between 1996 and 2003. Over this period, use doubled despite being associated with a higher mortality. However, the incidence of death decreased over time. Another example of surveillance of a new technology and assessment of associated outcomes is the Vermont Oxford Network of early nasal CPAP (continuous positive airways pressure).^[49] This study reports lower rates of chronic lung disease and retinopathy of prematurity associated with CPAP and suggests this may be due to early respiratory management. In this way, novel forms of analysis may be considered as hypothesis generating research. Such data could be used to generate interest in a large randomised multi-centre trial.

Conclusion
In order to maximise the potential health gains from a health informatics approach in neonatology it would be necessary to resolve a number of issues related to the ongoing collection, storage and analysis of data. These issues continue to present several challenges, particularly in terms of storage volume and refining outputs.

While automated analysis of the collected data is unlikely to be used as the sole basis for a clinical decision it can be used to inform that decision and the evidence suggests that the use of larger datasets along with improved analysis techniques ensure an enhanced predictive ability. Moreover, in the words of Charles Babbage "Errors using inadequate data are much less than those using no data at all". Potential barriers that would need to be addressed in order to increase utilisation are similar to those in other fields of health informatics. These include alleviation of manual data entry or labour intensive scanning procedures and, in some cases, difficulty with coding of specific neonatal clinical conditions.

However, there are also some aspects of neonatology that may act to facilitate improvements in data management. Particularly, neonatal networks that collect and pool data are an established and well-accepted concept with proven utility. Data submitted to a network undergoes a number of steps, including both local and central checking, to ensure integrity.

Further, the participating institutions are used to the collaborative methodology and receive summary data from the whole network. Increased data storage capacity and computer power are likely to be associated with further increases in the data warehousing and mining abilities available to interrogate data.

Finally, neonatology is by nature a technology friendly specialty. However, issues around data privacy need to be handled with both intelligence and sensitivity. Notwithstanding these issues, it seems reasonable to consider that health informatics is perhaps an underutilised tool in neonatology and there is potential to both expand use and increase utility. Hopefully, this opportunity to improve care and outcomes will be pursued and the learning experiences from this very specialised area of clinical practice will serve to inform data collection and analysis in other areas of medicine.

References

1. Neonatology. http://en.wikipedia.org/wiki/Neonatology. Accessed 22 March 2007.
2. Guidelines for Admissions to NICU, and Discharges and Transfers from NICU. http://www.adhb.govt.nz/newborn/Guidelines/Admission/NICUAdmissionsDischargesAndTransfers.htm. Accessed 22 March 2007.
3. de Courcy-Wheeler RH, Wolfe CD, Fitzgerald A, Spencer M, Goodman JD, Gamsu HR. Use of the CRIB (clinical risk index for babies) score in prediction of neonatal mortality and morbidity. Arch Dis Child Fetal Neonatal Ed. 1995;73(1):F32-6.
4. Parry G, Tucker J, Tarnow-Mordi W. UK Neonatal Staffing Study Collaborative Group CRIB II: an update of the clinical risk index for babies score. Lancet. 2003;361(9371):1789–91.
5. Kumar D, Super DM, Fajardo RA, Stork EE, Moore JJ, Saker FA. Predicting outcome in neonatal hypoxic respiratory failure with the score for neonatal acute physiology (SNAP) and highest oxygen index (OI) in the first 24 hours of admission. J Perinatol. 2004;24(6):376-81.
6. Snowden S, Brownlee KG, Dear PR. An expert system to assist neonatal intensive care. J Med Eng Technol. 1997;21(2):67–73.
7. Michnikowski M, Rudowski R, Siugocki P, Grabowski J, Rondio Z. Evaluation of the expert system for respiratory therapy of newborns on archival data. Int J Artif Organs. 1997;20(12):678–80.
8. Heermann LK, Thompson CB. Prototype expert system to assist with the stabilization of neonates prior to transport. Proc AMIA Annu Fall Symp. 1997;213–7.
9. Stoll BJ, Hansen N, Fanaroff AA, Wright LL, Carlo WA, et al. Changes in pathogens causing early-onset sepsis in very-low-birth-weight infants. N Engl J Med. 2002 ;347(4):240–7.
10. Jones B, Peake K, Morris AJ, McCowan LM, Battin MR. Escherichia coli: a growing problem in early onset neonatal sepsis. Aust N Z J Obstet Gynaecol. 2004;44(6):558–61.
11. Ng PC, Leung CW, Chiu WK, Wong SF, Hon EK. SARS in newborns and children. Biol Neonate. 2004;85(4):293–8.
12. O’Leary DR, Kuhn S, Kniss KL, Hinckley AF, Rasmussen SA, et al. Birth outcomes following West Nile Virus infection of pregnant women in the United States: 2003-2004. Pediatrics. 2006;117(3):e537–45.
13. Rolland E, Moore KM, Robinson VA, McGuinness D. Using Ontario’s "Telehealth" health telephone helpline as an early-warning system: a study protocol. BMC Health Serv Res. 2006;6:10.
14. Chan GC, Koh D. Reviewing lessons learnt of SARS in Singapore during planning for influenza pandemic. Int Marit Health. 2006;57(1-4):163–76.
15. Hand DJ. Mining medical data. Stat Methods Med Res. 2000;9(4):305–7.
16. Dybowski R, Weller P, Chang R, Gant V.Prediction of outcome in critically ill patients using artificial neural network synthesised by genetic algorithm. Lancet. 1996;347(9009):1146–50.
17. Walczak S, Scorpio RJ. Predicting pediatric length of stay and acuity of care in the first ten minutes with artificial neural networks. Pediatr Crit Care Med. 2000;1(1):42–7.
18. Livingstone DJ, Manallack DT, Tetko IV. Data modelling with neural networks: advantages and limitations. J Comput Aided Mol Des. 1997;11(2):135–42.
19. Sargent DJ. Comparison of artificial neural networks with other statistical approaches: results from medical data sets. Cancer. 2001;91:1636–42.
20. Almeida JS. Predictive non-linear modeling of complex data by artificial neural networks. Curr Opin Biotechnol. 2002 Feb;13(1):72–6.
21. Catley C, Frize M, Walker CR, Petriu DC. Predicting high-risk preterm birth using artificial neural networks. IEEE Trans Inf Technol Biomed. 2006;10(3):540–9.
22. Zernikow B, Holtmannspoetter K, Michel E, Theilhaber M, Pielemeier W, Hennecke KH. Artificial neural network for predicting intracranial haemorrhage in preterm neonates. Acta Paediatr. 1998;87(9):969–75.
23. Frize M, Ibrahim D, Seker H, Walker R, Odetayo MO, Petrovic D, Naguib RN. Predicting clinical outcomes for newborns using two artificial intelligence approaches. Conf Proc IEEE Eng Med Biol Soc. 2004;5:3202–5.
24. Ambalavanan N, Carlo WA. Comparison of the prediction of extremely low birth weight neonatal mortality by regression analysis and by neural networks. Early Hum Dev. 2001;65(2):123–37.
25. Zernikow B, Holtmannspoetter K, Michel E, Pielemeier W, Hornschuh F, Westermann A, Hennecke KH. Artificial neural network for risk assessment in preterm neonates. Arch Dis Child Fetal Neonatal Ed. 1998;79(2):F129–34.
26. Zernikow B, Holtmannspotter K, Michel E, Hornschuh F, Groote K, Hennecke KH. Predicting length-of-stay in preterm neonates. Eur J Pediatr. 1999;158(1):59–62.
27. Zernikow B, Holtmannspoetter K, Michel E, Theilhaber M, Pielemeier W, Hennecke KH. Artificial neural network for predicting intracranial haemorrhage in preterm neonates. Acta Paediatr. 1998;87(9):969–75.
28. Scope study. http://www.scopestudy.net/. Accessed 22 March 2007.
29. Ambalavanan N, Carlo WA, Bobashev G, Mathias E, Liu B, Poole K, Fanaroff AA, Stoll BJ, Ehrenkranz R, Wright LL; National Institute of Child Health and Human Development Neonatal Research Network. Prediction of death for extremely low birth weight neonates. Pediatrics. 2005;116(6):1367–73.
30. Mueller M, Wagner CL, Annibale DJ, Knapp RG, Hulsey TC, Almeida JS. Parameter selection for and implementation of a web-based decision-support tool to predict extubation outcome in premature infants. BMC Med Inform Decis Mak. 2006;6:11.
31. Thakkar M, O’Shea M. The role of neonatal networks. Semin Fetal Neonatal Med. 2006;11(2):105–10.
32. Simpson JM, Evans N, Gibberd RW, Heuchan AM, Henderson-Smart DJ; Australian and New Zealand Neonatal Network. Analysing differences in clinical outcomes between hospitals. Qual Saf Health Care. 2003;12(4):257–62.
33. Investigators of the Vermont Oxford Network. Horbar JD, Rogowski J, Plsek PE, et al. Collaborative quality improvement for neonatal intensive care. NIC/Q Project. Pediatrics 2001;107(1):14–22.
34. Synnes AR, Macnab YC, Qiu Z, Ohlsson A, Gustafson P, Dean CB, Lee SK; the Canadian Neonatal Network. Neonatal intensive care unit characteristics affect the incidence of severe intraventricular hemorrhage. Med Care. 2006;44(8):754–9.
35. Parry GJ, Gould CR, McCabe CJ, Tarnow-Mordi WO. Annual league tables of mortality in neonatal intensive care units: longitudinal study. International Neonatal Network and the Scottish Neonatal Consultants and Nurses Collaborative Study Group. BMJ. 1998;316(7149):1931–5.
36. Chien LY, Macnab Y, Aziz K, Andrews W, McMillan DD, Lee SK; Canadian Neonatal Network Variations in central venous catheter-related infection risks among Canadian neonatal intensive care units. Pediatr Infect Dis J. 2002 ;21(6):505–11.
37. Shah PS, Shah V, Qiu Z, Ohlsson A, Lee SK; Canadian Neonatal Network. Improved outcomes of outborn preterm infants if admitted to perinatal centers versus freestanding pediatric hospitals. J Pediatr. 2005;146(5):626–31.
38. Lee SK, Lee DS, Andrews WL, Baboolal R, Pendray M, Stewart S; Canadian Neonatal Network. Higher mortality rates among inborn infants admitted to neonatal intensive care units at night. J Pediatr. 2003;143(5):592–7.
39. Heuchan AM, Evans N, Henderson Smart DJ, Simpson JM. Perinatal risk factors for major intraventricular haemorrhage in the Australian and New Zealand Neonatal Network, 1995-97. Arch Dis Child Fetal Neonatal Ed. 2002 ;86(2):F86–90.
40. Henderson-Smart DJ, Hutchinson JL, Donoghue DA, Evans NJ, Simpson JM, Wright I; Australian and New Zealand Neonatal Network. Prenatal predictors of chronic lung disease in very preterm infants. Arch Dis Child Fetal Neonatal Ed. 2006;91(1):F40–5.
41. Darlow BA, Hutchinson JL, Henderson-Smart DJ, Donoghue DA, Simpson JM, Evans NJ; Australian and New Zealand Neonatal Network. Prenatal risk factors for severe retinopathy of prematurity among very preterm infants of the Australian and New Zealand Neonatal Network. Pediatrics. 2005;115(4):990–6.
42. Evans N, Hutchinson J, Simpson JM, Donoghue D, Darlow B, Henderson-Smart D. Prenatal predictors of mortality in very preterm infants cared for in the Australian and New Zealand Neonatal Network. Arch Dis Child Fetal Neonatal Ed. 2007;92(1):F34–40.
43. Aher SM, Ohlsson A. Early versus late erythropoietin for preventing red blood cell transfusion in preterm and/or low birth weight infants. Cochrane Database Syst Rev. 2006 Jul 19;3:CD004865.
44. Misra T, Dattani N, Majeed A. Congenital anomaly surveillance in England and Wales. Public Health 2006; 120:256–64.
45. McBride WG. Thalidomide and congenital abnormalities. Lancet 1962;2:1358. For full letter see : http://www.jameslindlibrary.org/trial_records/20th_Century/1960s/mcbride/mcbride_whole.html. Accessed 22 March 2007.
46. Rova M, Haataja R, Marttila R, Ollikainen V, Tammela O, Hallman M. Data mining and multiparameter analysis of lung surfactant protein genes in bronchopulmonary dysplasia. Hum Mol Genet. 2004;13(11):1095–104.
47. Kusiak A, Caldarone CA, Kelleher MD, Lamb FS, Persoon TJ, Burns A. Hypoplastic left heart syndrome: knowledge discovery with a data mining approach. Comput Biol Med. 2006;36(1):21–40.
48. Tingay DG, Mills JF, Morley CJ, Pellicano A, Dargaville PA; Australian and New Zealand Neonatal Network. Trends in use and outcome of newborn infants treated with high frequency ventilation in Australia and New Zealand, 1996-2003. J Paediatr Child Health. 2007;43(3):160–6.
49. Kirchner L, Weninger M, Unterasinger L, Birnbacher R, Hayde M, Krepler , Pollak A. Is the use of early nasal CPAP associated with lower rates of chronic lung disease and retinopathy of prematurity? Nine years of experience with the Vermont Oxford Neonatal Network. J Perinat Med. 2005;33(1):60–6.

Acknowledgements
I wish to thank Professor Jim Warren, Chair in Health Informatics, School of Population Health, University of Auckland and Ms Karen Day from Health Informatics, School of Population Health, University of Auckland for helpful comments and encouragement to submit this paper.